Re: heimdal -> MIT kdc migration

On Sat, Aug 30, 2025 at 4:22 PM Rick Macklem <rick.macklem@gmail.com> wrote:
>
> On Sat, Aug 30, 2025 at 8:56 AM Rick Macklem <rick.macklem@gmail.com> wrote:
> >
> > On Fri, Aug 29, 2025 at 1:05 PM Rick Macklem <rick.macklem@gmail.com> wrote:
> > >
> > > On Fri, Aug 29, 2025 at 7:43 AM Rick Macklem <rick.macklem@gmail.com> wrote:
> > > >
> > > > On Wed, Aug 27, 2025 at 8:39 PM Rick Macklem <rick.macklem@gmail.com> wrote:
> > > > >
> > > > > On Wed, Aug 27, 2025 at 7:43 PM Rick Macklem <rick.macklem@gmail.com> wrote:
> > > > > >
> > > > > > On Tue, Aug 26, 2025 at 9:35 AM Gleb Smirnoff <glebius@freebsd.org> wrote:
> > > > > > >
> > > > > > > On Tue, Aug 26, 2025 at 08:31:13AM -0700, Gleb Smirnoff wrote:
> > > > > > > T> On Tue, Aug 26, 2025 at 08:13:26AM -0700, Rick Macklem wrote:
> > > > > > > T> R> Ok. If you install FreeBSD-13.5 and then "pkg install heimdal", you get a
> > > > > > > T> R> working Heimdal-7.8 in ports.
> > > > > > > T> R>
> > > > > > > T> R> Now, I have another challenge. Fixing the master passwords.
> > > > > > > T> R> I'll work on it later to-day.
> > > > > > > T>
> > > > > > > T> I have applied two commits from Heimdal from 2012 that add 'kadmin dump -f MIT'
> > > > > > > T> feature to our base heimdal and polished them to compile.  So far it doesn't
> > > > > > > T> work yet, either create an empty dump or create a core dump, instead of
> > > > > > > T> database dump :) I'll see how difficult it is going to further resolve that to
> > > > > > > T> a working condition. If I succeed, then having 'dump -f MIT' in base without
> > > > > > > T> any ports would be the best solution.  Can also be merged to FreeBSD 14.4.
> > > > > > >
> > > > > > > Good news.  In the above paragraph I was testing my change incorrectly - threw
> > > > > > > the new binary on a system running unpatched libraries.  When run correctly,
> > > > > > > it successfully produced something that looks like a correct dump in MIT format.
> > > > > > > I haven't yet tried to load it into MIT kdc yet, though.
> Well, would you like the not so bad news or the bad news??;-)
> Your patch works, in that it produces a dump that "kdb5_util load
> -update" can load.
> After loading, if the principal only has keys for the newer encryption types of
> aes256-cts-hmac-sha1-96
> aes128-cts-hmac-sha1-96
> then you can look at the principal via kadmin.local, but the password must
> be changed before it works.
> --> This is the same behaviour as you get if you use Heimdal-7.8 to do the
>       dump conversion.
> So far, so good...
>
> Now, the not so good news. Once you update the Heimdal libraries
> (libhdb.so and libkadm5srv.so) "kadmin -l" is broken on the system
> running the old KDC. "kadmin -l dump" works, but something like:
> # kadmin -l
> kadmin> get rmacklem
> kadmin: get rmacklem: Service key not available
> - I have not yet looked in your patched sources to see where this
>   failure comes from?
>
> Now, more not so good news...
> My patch doesn't help.
> It does re-encrypt the key in the master key from the MIT KDC
> system, but that doesn't make the password work.
> When I compared the dump generated via kadmin with both
> your patch and mine, the key for aes256-cts-hmac-sha1-96
> is 34bytes long.
> After doing the change_password so that it works, a dump
> generated by "kdb5_util dump -r13" (the same dump format)
> has a key that is 62bytes long.
> --> So, there is more to converting the key than just re-ecrypting
>       it. (I'll try and find where the MIT code encrypts a key in a master
>       key to see why it ends up at 62bytes and whether that can be done
>       in the old code.)
>
> So, if we are going to continue with this...
> - We need to figure out why your patch breaks "kadmin" for other
>   things and fix that.
> - I/we need to figure out how to convert the 34byte key to the MIT
>   62byte key (and then maybe the password won't need to be changed?).
>
> Or do we just say "When you convert the KDC database, all the passwords
> must be changed to get them to work?".
All I've got sofar is this patch...
https://people.freebsd.org/~rmacklem/print.patch

It tweaks entry2mit_string_int() so that it skips over the keys for
old encryption types and fills in a fake "modified by" entry if none
exists.

These changes at least make the MIT dump such that the records
don't end up "incomplete or corrupted" when you try to do something
like "get_principal <principal>" in kadmin.local.

As noted, your patch makes "kadmin -l" break for most things,
reporting "Service key not available". The failures go away if
you revert back to the non-patched libraries.
I have not located the problem yet.

As for the passwords...no luck yet, rick


>
> rick
>
> > > > > Oh, and one more thing...
> > > > > - If there are keys for old encryption types like des.. or arcfour..
> > > > > in the MIT dump,
> > > > >   those will screw up the load. (You can check and delete them in the
> > > > > Heimdal-1.5.2
> > > > >   kdc system via..
> > > > >   # kadmin -l
> > > > >   get <principal>
> > > > >   - if old keys are listed in Keytypes:
> > > > >   del_enctype <principal> <enctypes>
> > > > >  exit
> > > > >
> > > > >  Ideally the conversion code would skip over these and not put them in the dump.
> > > > >
> > > > > rick
> > > > > ps: If you don't do this, when you "get_principal" in kadmin.local on
> > > > > the MIT KDC
> > > > >       system, it will give you a "Database record is incomplete or corrupted..".
> > > > >
> > > > > > >
> > > > > > > I will finalize the branch promptly and share it.  The above experience also
> > > > > > > indicated that I need to do a library version bump.
> > > > > > I don't know if you are enthusiastic about pursuing this, but hopefully this
> > > > > > works and gets the principals in (although I doubt the passwords will
> > > > > > work without changing them).
> > > > > >
> > > > > > To get the passwords to work, I think the following *might* do it:
> > > > > > - If you look in the Heimdal sources, when "--decrypt" is specified,
> > > > > >   I think it finds its way down into a function called hdb_unseal_key_mkey()
> > > > > >   which decrypts the key using the master key by calling _hdb_mkey_decrypt().
> > > > > >   To get the passwords to work, I think the call to _hdb_mkey_decrypt() would
> > > > > >   need to be followed by a call to _hdb_mkey_encrypt() with the "key"
> > > > > >   argument being the master key for the MIT database. (It it a keytab
> > > > > >   entry called /var/db/krb5kdc/.k5.YOUR.REALM created when you do a
> > > > > >   "kdb5_util create -s" on the system that will be the MIT KDC.)
> > > > > >   - Just to make it even more fun, there is a flag called HDB_KU_MKEY
> > > > > >     which is set to the Heimdal way and not for the MIT way (whatever
> > > > > >     that really means?).
> > > > > >   - There is also some stuff about padding in hdb_unseal_key_mkey(),
> > > > > >     but hopefully that won't be a problem?
> > > > > >
> > > > > > I think hdb_read_master_key() can be used to read in the MIT master
> > > > > > key from the file you provide as an argument to it.
> > > > > >
> > > > > > This all is just a hunch, based on what I've seen so far.
> > > > > >
> > > > > > I'll admit since the hardware I have takes forever to "make buildworld"
> > > > > > and I don't know a quick way to build/test these changes, I'm not
> > > > > > inspired to try it.
> > > > Although not inspired, I have taken a stab at it.
> > > > I am still trying to figure out how to build/test it, but I have forked
> > > > glebius@'s github and added some code to...
> > > > - Not dump the weak encryption keys (they just cause MIT's kadmin.local
> > > >   to complain that the principal's database entry is corrupted.
> > > > - If the argument to "kadmin -l dump" is "-f <mit-master-key-file>" instead
> > > >   of "-f MIT" it re-encrypts the keys in MIT's master key. (I hope that will
> > > >   make the passwords work.
> > > >   (Basically, someone will "kdb5_util create -s" on the MIT KDC system
> > > >    and then copy the /var/db/krbkdc/.k5.YOUR.REALM file over to the
> > > >    Heimdal KDC system and do "kadmin -l dump -f <.k5-filename> mit.dump"
> > > >    then copy "mit.dump" over to the MIT KDC system and
> > > >    "kdb5_util load -update mit.dump". Then, hopefully, the principals will
> > > >    work??)
> > > >
> > > > Anyhow, it is currently sitting here:
> > > > github.com/rmacklem/FreeBSD in the kadmin-dump-MIT branch.
> > > > (I'm as unconversant with git and github as anyone, so if you have
> > > > trouble finding it, just let me know.)
> > > Actually, it hasn't made it there yet. For some reason (I think it is
> > > glebius@s large # of branches) it takes a very long time to "git push"
> > > a patch involving 4 files. It failed after over an hour with an unexpected
> > > TCP disconnect. I am running it again.
> > >
> > > I will stick the patch here, in case the push fails again.
> > > (It needs to be applied on top of glebius@'s kadmin-dump-MIT branch.
> > The patch is here. (For some reason, I couldn't push so I deleted the
> > github fork.)
> > https://people.freebsd.org/~rmacklem/kadmin.patch
> >
> > I haven't yet been able to test it, but will be able to do so later to-day, rick
> >
> > >
> > > Meanwhile I've given up trying to build it on a universe system and
> > > an now trying the "make buildworld" locally. This will take days,
> > > so I guess I'll go do something else.;-)
> > >
> > > rick
> > >
> > > >
> > > > I'll keep updating this github fork as I get to test it, but if others
> > > > know how to build it, feel free to test, rick
> > > >
> > > > > >
> > > > > > rick
> > > > > >
> > > > > > >
> > > > > > > --
> > > > > > > Gleb Smirnoff